Y Combinator Partner: Transforming 23 Years of Email and Calendar Data into gbrain Memory Bank
Y Combinator partner Tom Blomfield stated that he is preparing to import 23 years of his Gmail, Calendar, and Evernote data into gbrain and convert it into vector embeddings. In other words, he wants to turn his long-accumulated personal information into a retrievable, callable AI memory bank.
This approach is becoming an important direction for AI personal assistant products. English materials indicate that more and more tools are beginning to unify the vectorization of emails, calendars, notes, and chat records to support long-term memory, semantic search, and task tracking, aiming to enable AI not only to "answer questions" but also to "know who you are and what you have done."
Blomfield's actions have attracted attention because they represent a shift in AI applications from general Q&A to personal data operating systems. The real value is no longer just in how smart the model is, but in whether it can capture users' decades of behavioral trajectories and turn these fragments into continuously usable context.
Source: Public Information
ABAB AI Insight
Blomfield's statement centers not on "organizing data" but on the structural reconstruction of personal cognitive assets. The 23 years of emails, calendars, and notes are not static archives but represent a person's decision history, relationship networks, work rhythms, and evolving interests. Converting them into vector embeddings essentially establishes a long-term memory layer for AI.
This reflects that AI applications are entering a deeper stage: moving from answering questions to understanding users. Past assistants could only see current inputs; now, if assistants can access historical context, they can better predict intent, fill in gaps, and reduce repetitive work. The real leap is not in model parameters but in memory and retrieval structures.
On a deeper level, this trend will make "personal data integration capability" a new barrier. Whoever can unify information scattered across emails, calendars, documents, and chat software will be closer to becoming the default entry point for users' work. Future competition will not only be about how good the model is but also about who controls your work traces from the past twenty years.
From an industrial structure perspective, this is also a signal of AI evolving from "tools" to "operating systems." Models are responsible for reasoning, embeddings for memory, and products for invocation, ultimately forming a closed loop centered around users' lives and work. What Blomfield is doing is essentially creating a prototype for this closed loop.